Using Boosted Regression to Understand Irrigation Decision-Making
Abstract
Groundwater levels across parts of western Kansas have been declining at unsustainable rates due to pumping for agricultural irrigation despite technological advances designed to decrease total water use from the underlying aquifer (e.g., efficient irrigation, drought-resistant cultivars). Thus, water management across this agricultural landscape is more complex than targeting a simple water budget. Instead, both qualitative (e.g., management boundaries) and quantitative (e.g., crop prices) factors drive unsustainable water applications. This study uses a boosted regression tree machine-learning technique to simultaneously analyze categorical and numerical data against annual irrigation pumping. We test ~40 key variables to irrigation use from 1996-2017 to derive predictor coefficients of groundwater pumping for each variable. The results allow for distinguished influence of various drivers such as governing policies, crop type, or soil characteristics on total irrigation pumping across both space and time. By targeting the combinations of factors that statistically lead to the greatest volumes of groundwater pumping, robust management strategies can be developed to achieve conservation goals adopted in the region. This statistical approach can be replicated for other large-scale studies across the country, particularly in regions seeking to better understand irrigation use in a changing agricultural landscape.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2019
- Bibcode:
- 2019AGUFM.H13M1901L
- Keywords:
-
- 0402 Agricultural systems;
- BIOGEOSCIENCES;
- 1842 Irrigation;
- HYDROLOGY;
- 1880 Water management;
- HYDROLOGY;
- 6309 Decision making under uncertainty;
- POLICY SCIENCES